| With the continuous rise of vehicle ownership,traffic safety has attracted more and more attention from all walks of life.Assistant driving system is one of the key research objects in automobile production industry in recent years.A well-functioning auxiliary system can not only bring safe and good driving experience to drivers,but also effectively ensure the safety of public transport.Among the many functions of auxiliary driving system,lane line detection system and traffic sign recognition system are key research topics.A well-functioning auxiliary system can not only bring safe and good driving experience to drivers,but also effectively ensure the safety of public transport.The research content of this paper is mainly divided into two parts:First is to locate and recognize traffic signs in the process of driving,and the second is to recognize and extract lane lines according to the collected pavement information.In the aspect of traffic sign recognition,this paper uses the deep learning method to locate and recognize the traffic signs.While analyzing the yolo-v5 s network structure,it improves some structures,and puts forward a joint neural network model.On the basis of locating the traffic signs in the image by using yolo-v5 s network,it uses some structures of VGGNet to recognize the image,FPN-PAN structure is adopted to deepen the recognition ability of the network to small targets.At the same time,the global average pooling layer is used to replace the full connection layer of the traditional VGGNet for output,so as to reduce the huge number of parameters brought by the full connection layer and improve the network operation efficiency.In the aspect of lane line extraction,a lane line information extraction algorithm based on Gabor filter is proposed.Before the detection,the algorithm first uses HSI color model to enhance the image data,so as to improve the detection ability of the algorithm to the worn lane line.In the detection process,firstly,the vanishing points of the driving road are searched by using the texture directional characteristics of the pavement,and then the road boundary is searched iteratively and the vanishing points are updated based on the first extracted vanishing points.After extracting the area where the road is located,the road area is converted from the vehicle driving angle to the top view angle by using the inverse transmission transformation.After the noise reduction of the road area after the inverse transmission transformation,the single lane line area is extracted and transformed.This paper uses the fitting algorithm based on Catmull-ROM spline to fit the lane line,and samples the control points from the center line of each lane area to improve the fitting degree of the algorithm to the curved lane line.Finally,the paper integrates the above two parts of the software,and uses the established test set to test the experiment.Experiments show that in the part of traffic sign recognition,the accuracy of the joint neural network used in this paper is 4.21% higher than that of the traditional VGGNet,2.08% higher than that of the yolo-v5 s network,and the detection effect of small targets is 2.66% higher than that of the yolo-v5 s network.In terms of lane line detection,through the data set test,the false detection rate is 1.9%,which is 0.9% higher than the existing RANSAC lane line fitting algorithm.The algorithm used in this paper is feasible. |
